Hashing Forests for Morphological Search and Retrieval in Neuroscientific Image Databases
In this paper, for the first time, we propose a data-driven search and retrieval (hashing) technique for large neuron image databases. The presented method is established upon hashing forests, where multiple unsupervised random trees are used to encode ne
- PDF / 824,279 Bytes
- 9 Pages / 439.363 x 666.131 pts Page_size
- 58 Downloads / 208 Views
mputer Aided Medical Procedures, Technische Universit¨ at M¨ unchen, Germany 2 Max Plank Digital Library, M¨ unchen, Germany Computational Neuroscience, Ludwigs Maximillian Universit¨ at M¨ unchen, Germany 4 Computer Aided Medical Procedures, Johns Hopkins University, USA
Abstract. In this paper, for the first time, we propose a data-driven search and retrieval (hashing) technique for large neuron image databases. The presented method is established upon hashing forests, where multiple unsupervised random trees are used to encode neurons by parsing the neuromorphological feature space into balanced subspaces. We introduce an inverse coding formulation for retrieval of relevant neurons to effectively mitigate the need for pairwise comparisons across the database. Experimental validations show the superiority of our proposed technique over the state-of-the art methods, in terms of precision-recall trade off for a particular code size. This demonstrates the potential of this approach for effective morphology preserving encoding and retrieval in large neuron databases.
1
Introduction
Neuroscientists often analyze the 3D neuromorphology of neurons to understand neuronal network connectivity and how neural information is processed for evaluating brain functionality [1, 2]. Of late, there is a deluge of publicly available neuroscientific databases (especially 3D digital reconstructions of neurons), which consist of heterogeneous multi-institutional neuron collection, acquired from different species, brain regions, and experimental settings [3, 4]. Finding relevant neurons within such databases is important for comparative morphological analyses which are used to study age related changes and the relationship between structure and function [1]. Recently, the concept of the neuromorphological space has been introduced where each neuron is represented by a set of morphological and topological measures [1]. Authors in [1] used this feature space for multidimensional analysis of neuronal shape and showed that the cells of the same brain regions, types, or species tend to cluster together in such a space. This motivated us to leverage this space for evaluating inter-neuron similarity and propose a data-driven retrieval (hashing) system for large neuron databases. Recently, a neuron search ∗
S. Mesbah and S. Conjeti contributed equally towards the work.
c Springer International Publishing Switzerland 2015 N. Navab et al. (Eds.): MICCAI 2015, Part II, LNCS 9350, pp. 135–143, 2015. DOI: 10.1007/978-3-319-24571-3_17
136
S. Mesbah et al.
algorithm was proposed by [2], where pairwise 3D structural alignment was employed to find similar neurons. In another approach, authors in [5] focused on the evaluation of morphological similarities and dissimilarities between groups of neurons deploying clustering technique using expert-labeled metadata (like species, brain region, cell type, and archive). These existing neuronal search and retrieval systems are either too broad (lacking specificity in search) or too restrictive (dependent on ex
Data Loading...
